ftp://ftp.fwi.uva.nl/pub/theory/illc/researchreports/DS-95-14.text.ps.gz
or with a link from:
http://www.fwi.uva.nl/research/illc/wwwreports.html
Rens Bod
(Univ. of Amsterdam, NL)
ABSTRACT:
The book starts by motivating a statistical approach to linguistics from
both a cognitive and an engineering point of view, and pursues with the
problem of what should be demanded from a statistical enrichment of a
given linguistic theory. It then argues for a linguistic performance
model which employs a very large language corpus, standing for a person's
past language experience, in which each sentence is annotated with the
analysis that seemed most appropriate for understanding the sentence in
the context in which it was uttered. An analysis of a new sentence can be
constructed out of combinations of partial analyses that occur in the
corpus. By combining the relative frequencies of these partial analyses,
the model is able to select from all possible analyses of a sentence the
analysis which is actually perceived by a person. The book deals with six
different realizations of performance models, that allow for the use of
currently available corpora, and goes into their formal, computational
and experimental aspects.
=======================================================================